Waveform

A waveform is the most basic representation of audio — a plot of signal amplitude (air pressure variation) against time. Digital audio stores waveforms as sequences of numerical samples, with each sample representing the instantaneous amplitude at that point in time.

While waveforms contain all the information in an audio signal, they are not directly useful for speech recognition because the raw amplitude values do not clearly reveal the frequency content that distinguishes different speech sounds. This is why audio is typically converted to spectral representations (spectrograms, mel spectrograms, MFCCs) before being processed by ASR models.